As is well known, memory cells having a single floating gate transistor are most widely utilized in FLASH memories. To better illustrate the invention aspects, a conventional type of memory cell, as schematically shown in FIG. 1, will be considered first. Such a cell 1 comprises a MOS transistor 2 which is realized over a semiconductor substrate 3, and has a source region S, and a drain region D separated by a channel region C which, in turn, has a control gate terminal GC realized thereover. This is, therefore, a transistor of the NMOS type wherein a second gate GF, referred to as the floating gate on account of it being fully enveloped in a dielectric material, is also provided between the control gate GC and the substrate 3. In particular, the floating gate GF is separated from the control gate GC by a layer 4, referred to as the interpoly dielectric layer. A thin layer 5 of silicon oxide, referred to as the gate oxide layer, is interposed between the floating gate GF and the substrate 3.
Thus, the floating gate GF cannot be accessed from outside, and its potential will only depend on any charge which may be present thereon and on the capacitive couplings with the other elements of the cell. Accordingly, such a cell forms basically a data storage element based on charges becoming trapped on the floating gate GF. Major advantages of a floating gate memory cell are: its compact size, low sensitivity to disturbances during the programming step, and low voltage requirements during the reading step, that is, for both the programming operations and the cell erasing.
Such a memory cell 1 is usually programmed by hot electron injection and erased by a tunneling effect. The programming operation of a memory cell such as the cell 1 uses the generation of channel hot electrons in the drain region D (or rather, in an area of spatial charge thereof known as the pinch-off area). Through the application of a favorable electric field, i.e., a field which enables a certain number of electrons of the drain region D to acquire sufficient energy to overcome the potential barrier between the drain region and the gate oxide, charges are caused to migrate toward the floating gate GF. The floating gate GF will store an essentially negative charge.
In a dual mode, for the erasing operation, which allows electrons to be extracted from the floating gate GF of the memory cell 1, the tunneling effect is utilized, also referred to as Fowler-Nordheim's effect. In particular, current programming techniques provide for the erasing to be carried out by applying a high voltage to the source region S, this operation being commonly referred to as source erasing. Under these conditions, the shape of the source junction should be a gradual one, so that the substrate current can be lowered and the risk of creating hot holes reduced.
The need for a gradual source junction shape represents, however, one of the strictest limitations imposed on the scalability of the gate length of a memory cell so realized. A first known approach for overcoming this limitation is to perform the operation of erasing of the memory cell 1 by a so-called channel erasing. In this case, the memory cell 1 is erased by applying a negative voltage to the control gate GC of the transistor 2, while holding the source S and drain D regions at the same potential as the semiconductor substrate 3.
Channel erasing mitigates the limitations on the length scalability of the cell 1, thereby removing the need for a particular shape of the source junction. While being in many ways advantageous, this first discussed approach has a major drawback in that it involves the application of high erase voltages to the control gate GC, due to the high capacitive coupling between the channel and the floating gate. In fact, while for source erasing an erase voltage of 10 to 12 volts is adequate, channel erasing requires a voltage of 17 to 20 volts. Such a high voltage value is, however, incompatible with advanced CMOS processes for manufacturing semiconductor devices.
To address this problem, it has been prior practice to locate the storage device in a diffusion well of the P-well type, isolated from the substrate 3, using a triple-well CMOS process. In this way, positive voltages can be applied to the substrate 3 and the source and drain regions of the memory cell during the erasing operation, so as to split the voltage between the negatively biased control gate GC and the positively biased substrate 3. This would, accordingly, restrict the absolute value of the erase voltage needed.
With a view to obtaining a scalability for the length of the cell 1 it is necessary, however, to concurrently raise the inherent threshold voltage Vt0 of the memory cell 1 so as to attenuate the effects of a possible shorting of the channel region C (punch-through, drain-turn-on). The inherent threshold voltage Vt0 of the memory cell depends on the characteristics of the cell construction (such as the thickness of the gate oxide 5 or the capacitive ratios) and the dopant concentration in the channel region C.
The erase voltage required to have the cell erased within a predetermined time, usually set by the storage device specifications, will be higher, the higher the inherent threshold voltage Vt0 of the cell. In fact, the threshold voltage Vt of the cell depends on the charge Q stored in the floating gate GF, and the inherent threshold voltage Vt0 of the cell (that is, when it has no charge in the floating gate), according to the following relationship: EQU Vt=Vt0-Q/Cpp
being the coupling capacitance between the floating gate GF and the control gate GC.
In addition, for the memory reading operations to be carried out properly, it is necessary for the threshold voltages of the memory cells to acquire suitable values, regardless of their inherent threshold voltage Vt0. In particular, memory cells containing a logic state "1" should have a threshold voltage typically within the range of 0.5V to 3V, whereas memory cells containing a logic state "0" should have a threshold voltage typically above 5V.
As for memory cells of the FLASH type, current manufacturing processes provide for inherent threshold voltages between 2V and 3V. Thus, erased cells (i.e. cells with a logic state "1") will have a threshold voltage Vt close to their inherent threshold voltage Vt0, or in other words, have a reduced number of charges on the floating gate GF, whereas programmed cells (i.e. cells with a logic state "0") will show a threshold voltage Vt which is definitely higher than their inherent threshold voltage Vt0. In this context, the erasing operation includes extracting the negative charge stored in the floating gate GF of the programmed cells until it is restored to a condition of near-neutral charge.
In addition, the presence of charge in large amounts in the floating gate of the programmed cells contributes to an increased electric field in the gate oxide during erasing. As mentioned above, to obtain a scalability for the length of the cell 1, the dopant concentration in the channel region C must be increased so as to attenuate the punch-through and drain-turn-on effects. In this way, memory cells can be obtained which are scaled with inherent threshold voltages Vt0 of 4 to 6 volts.
However, such cells are in a near-neutral charge condition in the programmed state of "0", and have a high positive charge in the floating gate GF when in the erased state. This positive charge lowers the electric field in the channel oxide during the erasing operation and enforces the application of higher erase voltages. Thus it occurs that, while a negative erase voltage of 17-20V must be applied to the gate of a cell having an inherent threshold voltage Vt0 of 2-3V, in order to erase a memory cell with an inherent threshold voltage of 4-6V, a negative erase voltage of 20-24V must be provided. The need for scaling the cell length would, therefore, lead to the use of high erase voltages, which clashes with advanced CMOS processes that have a limited capability to handle high voltages.